โ† All examples

ModelEthio-ASR Amharic (600M)1Ethio-ASR Multi (300M)6Ethio-ASR Multi (600M)3Ethio-ASR Multi (94M)6Meta MMS-1B7OmniASR-CTC (1B)7OmniASR-CTC (300M)10OmniASR-CTC (3B)1OmniASR-LLM (1B)6OmniASR-LLM (300M)9OmniASR-LLM (3B)9OpenAI Whisper-small7
Audio qualityClean Studio3Ambient Room Noise2Harsh Environment4Distant / Low-Volume Mic4Wideband Phone (VoLTE)6Randomized Phone Channel6Narrowband 2G + Packet Loss7

What did the model get wrong?

Ethio-ASR Multi (94M) Randomized Phone Channel   Example 10 of 20  ยท  speaker: Male

๐ŸŽง Listen โ€” Randomized Phone Channel
Phone-call audio with random extra distortions layered on top.
โœ… What was actually said
แˆตแˆˆ แŒฅแ‰…แˆŽแ‰น แ‹แˆญแ‹แˆญ แˆ˜แˆจแŒƒ แ‹จแ‰ต แŠแ‹ แ‹จแˆ›แŒˆแŠ˜แ‹
๐Ÿค– What this model heard
amh แˆตแˆˆแ‰ตแˆŽแ‰น แ‹แ‹แˆญ แˆ˜แˆจแ‹ฃแ‹ญแ‰ต แŠแ‹ แˆ›แŒˆแŠ˜แ‹
Correct
Wrong (different word)
Missed (skipped)
Extra (added)
1
Correct
5
Wrong
1
Missed
0
Extra
14%
Words right
Out of 7 spoken words, this model got 1 right (14%). Errors: 5 words wrong; 1 missed.

Try switching Audio quality โ€” see the same model degrade as the audio gets noisier.